Preparing your design workspace

Loading the problem and your latest saved attempt…

Equity Research Decision Intelligence

Design a research platform that turns a point-in-time document corpus into auditable next-day equity decisions.

Problem statement

For this exercise, assume the source-backed system identity and the explicitly bounded practice contract that follow. Design an end-to-end equity research decision platform.

It ingests roughly ten years and four million research reports, news items, and earnings materials; reconstructs the information known about each company at a requested point in time; and produces an auditable next-day trading decision.

The operating team is one research engineer and one platform engineer, with eight B200-class accelerators available. Keep the design feasible for that team, distinguish research-time experimentation from the production decision path, and explicitly decide where open-source models or hosted APIs are appropriate.

What to build

  1. Ingest, normalize, and version research reports, news, and earnings materials without losing event time or publication time.
  2. Reconstruct a company view using only information that was available by the requested cutoff.
  3. Produce a next-day decision with the evidence, model version, features, and configuration needed to reproduce it.
  4. Let researchers evaluate candidate retrieval and inference changes without silently altering production history.

At scaleExercise boundary

  • The historical corpus spans about ten years and four million source documents.
  • Eight B200-class accelerators are available, so scheduling and isolation must prevent exploratory jobs from starving daily production work.
  • The system is operated by one research engineer and one platform engineer, favoring a small number of observable managed components.
  • A daily decision must use a point-in-time snapshot and finish before the next trading session.